Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Zebra-Llama – Towards efficient hybrid models

arxiv.org

Post reply on HN
Print view View original

61–66 of 66 posts

Re: Zebra-Llama – Towards efficient hybrid models

#61

Post by colechristensen »

buu700 wrote: post #60
nl wrote: post #59

Earlier quoted context omitted.

If you haven't tried these things since Sonnet 4.5 came out then it's time to give them another try. Sonnet 4.5 and especially Codex 5.1 have completely changed the way I build software. > The fact that tools like Replit also include their own hosting environments is definitely neat, but not really what I was getting at as far as deployment. What I had in mind was managing arbitrary cloud platforms, setting up an opt…

Just in case it was unclear, I extensively use AI and agentic coding with current models on a daily basis. The only thing I haven't tried in a few months is specifically one-shotting a greenfield project. I know computer-use agents exist, and theoretically have tooling and permission to do all the things a human sitting in front of a computer can. I just haven't heard of anyone successfully claiming to have had one d…

Well... they're not oracles and never will be. The things I'm creating are following recognizable development practices. It's not build-once and done, it's an elaborate design/build/test cycle that happens in many flavors because unless you've already done something and are copying it, that's how you create and language models aren't going to get away from that.

Re: Zebra-Llama – Towards efficient hybrid models

#62

Post by buu700 »

sureglymop wrote: post #49
buu700 wrote: post #34

Earlier quoted context omitted.

Coding capability in and of itself may be "good enough" or close to it, but there's a long way to go before AI can build and operate a product end-to-end. In fairness, a lot of the gap may be tooling. But the end state in my mind is telling an AI "build me XYZ", having it ask all the important questions over the course of a 30-minute chat while making reasonable decisions on all lower-level issues, then waking up the…

Given that natural language is ambiguous, what if the LLM makes some mistakes though? I'm wondering because, it's not like it's a human that can then take accountability/responsibility for that...

I'd say the simple answer is that the buck always stops with the vendor. If Acme Co sells a jetpack that explodes and kills someone, the entity doesn't get to deflect liability by saying one of their engineers made a mistake; that may be an explanation, but not an excuse. Swapping out Acme Co and its employees with your grandma and her AI/robots doesn't change the fundamental principle.

Re: Zebra-Llama – Towards efficient hybrid models

#63

Post by buu700 »

colechristensen wrote: post #61
buu700 wrote: post #60

Earlier quoted context omitted.

Just in case it was unclear, I extensively use AI and agentic coding with current models on a daily basis. The only thing I haven't tried in a few months is specifically one-shotting a greenfield project. I know computer-use agents exist, and theoretically have tooling and permission to do all the things a human sitting in front of a computer can. I just haven't heard of anyone successfully claiming to have had one d…

Well... they're not oracles and never will be. The things I'm creating are following recognizable development practices. It's not build-once and done, it's an elaborate design/build/test cycle that happens in many flavors because unless you've already done something and are copying it, that's how you create and language models aren't going to get away from that.

Whether or not it will one day get there is anyone's guess, but it sounds like we agree that it at least isn't currently there. I brought up that goalpost to illustrate why more efficient models will only improve the aggregate volume and/or quality of output for the foreseeable future, as opposed to creating a glut of supply that destroys the economics of data centers.

Re: Zebra-Llama – Towards efficient hybrid models

#64

Post by credit_guy »

jychang wrote: post #43
credit_guy wrote: post #36

Earlier quoted context omitted.

Here's what's important about this paper. It is written by AMD researchers. It shows AMD is investing in AI research. Is this the same level of achievement as DeepSeek 3.2. Most likely not. Do they have novel ideas? Difficult to say, there are hundreds of new ideas being tried in this space. Is this worthless? Most certainly not. In order to make progress in this domain (as in any other), you first need to get your f…

Mamba based LLMs aren't even close to novel though. IBM's been doing this since forever [1]. Also, you're off on Deepseek V3.2's param count, the full model's 685B in size with the MTP layer. I don't think there's anything interesting here other than "I guess AMD put out a research paper", and it's not cutting edge when Deepseek or even IBM is running laps around them. [1] Here's a news article from April, although I…

It's not cutting edge, so what? Your point is that nobody should publish anything unless it is cutting edge?

Re: Zebra-Llama – Towards efficient hybrid models

#65

Post by jychang »

credit_guy wrote: post #64
jychang wrote: post #43

Earlier quoted context omitted.

Mamba based LLMs aren't even close to novel though. IBM's been doing this since forever [1]. Also, you're off on Deepseek V3.2's param count, the full model's 685B in size with the MTP layer. I don't think there's anything interesting here other than "I guess AMD put out a research paper", and it's not cutting edge when Deepseek or even IBM is running laps around them. [1] Here's a news article from April, although I…

It's not cutting edge, so what? Your point is that nobody should publish anything unless it is cutting edge?

Yeah, that's the point of publishing. You get scooped, you lose.

Re: Zebra-Llama – Towards efficient hybrid models

#66

Post by credit_guy »

jychang wrote: post #65
credit_guy wrote: post #64

Earlier quoted context omitted.

It's not cutting edge, so what? Your point is that nobody should publish anything unless it is cutting edge?

Yeah, that's the point of publishing. You get scooped, you lose.

This wasn’t published, it was just posted to the arxiv.
Post reply on HN